High-grade brain gliomas, the most common of which is glioblastoma multiforme (GBM), have terrible prognosis and a median patient survival of about 12 months. Although combinations of surgical removal, radiotherapy and chemotherapy are used in the clinical practice, a fundamental and persistent limitation in treating these aggressive tumors is that they tend to infiltrate into normal tissue well beyond margins visible via imaging. Since assessing the spatial extent of tumor infiltration is nearly impossible using current radiologic reading practices, clinical treatment tends to be restricted to parts that are deemed to be clearly malignant, frequently only the enhancing tumor. This failure to aggressively treat the infiltrating tumor accelerates tumor recurrence, and eventually patient death. This proposal aims to develop computational modeling and image analysis methods that will improve our ability to estimate GBM infiltration, as well as to predict tissue that is likelyto present fastest tumor recurrence, thereby eventually opening the way for more aggressive, yet targeted, treatment, such as targeted aggressive surgical removal and/or radiosurgery. To achieve our goal, we will integrate information from several sources: 1) advanced multi-parametric imaging, which captures many aspects of tumor anatomy and physiology~ 2) computational modeling of tumor growth and infiltration~ 3) machine learning methods which, after appropriate training, can learn subtle and potentially complex imaging phenotypes of infiltrating tumors~ 4) statistical atlases, which capture population-based trends that can offer additional insights into tumor growth, such as relationship of infiltration to vasculature and to white matter fiber pathways~ 5) data from one of the largest patient populations having advanced imaging, genotyping, follow-up till tumor recurrence, and histological analysis.